Home Artificial Intelligence Researchers build synthetic dog dataset in GTA-V to create 3D models

Researchers build synthetic dog dataset in GTA-V to create 3D models

by Dimitrios Koutrotsios
dog ai

Researchers from the University of Surrey have developed a method to transform photographs of dogs into detailed 3D models. 

Researchers aimed to train an AI system to interpret and convert 2D images of dogs into their 3D poses.

The training material? Not real dogs, but rather computer-generated images from the virtual world of GTA V. 

Moira Shooter, a postgraduate research student involved in the study, shared, “Our model was trained on CGI dogs – but we were able to use it to make 3D skeletal models from photographs of real animals. That could let conservationists spot injured wildlife, or help artists create more realistic animals in the metaverse.”

The traditional methods for teaching AI about 3D structures involve using real photos alongside data about the objects’ actual 3D positions, often obtained through motion capture technology. 

However, when applying these techniques to dogs, there are simply too many movements to track.  

To build their dog dataset, researchers altered GTA V’s code to replace its human characters with dog avatars through a process known as “modding.” 

Dogs
Examples from the synthetic dog dataset generated using GTA-V. Source: University of Surrey.

This enabled them to produce 118 videos capturing these virtual dogs in various actions – sitting, walking, barking, and running – across different environmental conditions. This resulted in the creation of ‘DigiDogs,’ a rich database containing 27,900 frames of dog movement, captured in a way real-world data collection hadn’t allowed.

The next steps used Meta’s DINOv2 AI model for its strong generalization skills, fine-tuning it with DigiDogs to accurately predict 3D poses from single-view RGB images.

Researchers demonstrated that using the DigiDogs dataset for training resulted in more accurate and lifelike 3D dog poses than those trained on real-world datasets, thanks to the variety in dog appearances and actions captured.

AI dogs
Models trained on the synthetic dataset DigiDogs showed improved accuracy versus those trained only with the real-world dataset RGBD-Dogs. Source: University of Surrey.

The results surpassed existing methods by providing detailed 3D results and establishing a new benchmark in both realism and accuracy for 3D dog pose estimation from 2D images, confirmed through thorough qualitative and quantitative evaluations.

While this study represented a big step forward in 3D animal modeling, the team acknowledges there’s more work to be done, especially in improving how the model predicts the depth aspect of the images (the z-coordinate). 

Shooter described the potential impact of their work, saying, “3D poses contain so much more information than 2D photographs. From ecology to animation—this neat solution has so many possible uses.”

The paper won the Best Paper prize at the IEEE/CVF Winter Conference on Applications of Computer Vision but also promises many applications, from wildlife conservation to digital 3D object rendering in VR applications. 

Source Link

Related Posts

Leave a Comment